Frequency response functions (FRF) are the most fundamental data for the frequency domain identifications of the dynamic characteristics of structural systems. Several techniques have emerged for the estimation of FRF so as to minimize the effect of measurement noise. This paper presents an improved method for estimating FRFs. The new FRF estimator takes the weighted average of two conventional estimators, H1(f) and H2(f), utilizing the characteristics that H2(f) gives a more accurate estimate at resonance, while H1(f) yields a better result at antiresonance. An exponential weighting function is proposed, and the parameter of the weighting function is determined by minimizing the resolution bias error of the FRF over the significant frequency range. Based on the estimated FRFs, the modal parameters of the structural systems, such as, natural frequencies, damping ratios and mode shapes, are estimated. The effectiveness of the proposed method is investigated through numerical and experimental studies. The estimated results indicate that the proposed estimator gives more accurate results than other estimators.